Storage of sparse-coded hetero-associations with the competitive synaptic growth network

The authors apply the competitive synaptic growth network (SGN) to the storage and recall of sparse coded binary heteroassociations. The SGN is a two-layer, feedforward neural network utilizing complex neurons having finite extent and active dendritic structures. The performance measure for the SGN is the number of connections needed to store noiseless sparse coded heteroassociations and perfectly recall these associations when given a (perhaps noisy) input vector. Using the same number of network connections, the SGN clearly outperforms the benchmark nonholographic associative memory applied to the same problem.<
>